Echo Cancellation Using Neural Networks for Environments with Unsynchronized Devices for Audio Capture and Rendering

3. The system as recited in claim 1, wherein at least one neural network of the first and second neural networks comprises a recurrent neural network.

4. The system as recited in claim 1, wherein an activation function of at least one neural network of the first and second neural networks is a complex activation function.

5. The system as recited in claim 1, wherein a learned weight of at least one neural network of the first and second neural networks is expressed as a complex number.

7. The computer-implemented method as recited in claim 6, wherein at least one neural network of the first and second neural networks comprises a recurrent neural network.

8. The computer-implemented method as recited in claim 6, wherein at least one neural network of the first and second neural networks comprises one or more gated recurrent units (GRUs).

9. The computer-implemented method as recited in claim 6, wherein a learned weight of at least one neural network of the first and second neural networks is expressed as a complex number.

10. The computer-implemented method as recited in claim 6, wherein an activation function utilized at a neural network of the first and second neural networks comprises one of: (a) a cardioid function or (b) a siglog function.

11. The computer-implemented method as recited in claim 6, wherein a neural network of the first and second neural networks comprises a layer at which a convolution operation is performed with respect to a representation of a reference signal obtained from the second communication environment.

12. The computer-implemented method as recited in claim 6, wherein the first set of one or more properties for which a compensation is applied at the first layer includes a non-linearity in audio reproduction of the first audio rendering device.

14. The computer-implemented method as recited in claim 6, wherein the first audio capture device is incorporated within a voice-driven personal assistant device.

17. The one or more non-transitory computer-accessible storage media as recited in claim 16, wherein at least one neural network of the first and second neural networks comprises a recurrent neural network.

18. The one or more non-transitory computer-accessible storage media as recited in claim 16, wherein a learned weight of at least one neural network of the first and second neural networks is expressed as a complex number.

19. The one or more non-transitory computer-accessible storage media as recited in claim 16, wherein the first set of one or more properties for which a compensation is applied at the first layer includes a non-linearity in audio reproduction of the first audio rendering device.